PHY PMA electrical specs baseline proposal for 803.an

Transcription

1 PHY PMA electrical specs baseline proposal for 803.an Sandeep Gupta, Teranetics Supported by: Takeshi Nagahori, NEC electronics Vivek Telang, Vitesse Semiconductor Joseph Babanezhad, Plato Labs Yuji Kasai, AIST 7/13/2004 1

2 Basic Approach taken In 1000BASE-T PMA electrical specs are addressed by: Transmit waveform test Transmit droop test Transmit distortion test Transmit jitter test Master mode Slave mode MDI impedance balance test MDI Common mode output voltage test To get more complete PMA electrical specs for 10GBASE-T, the following must be specified Modulation, if PAM, how many levels? Symbol rate current proposals range from 810MHz to 1000MHz Filtering of the Transmitter digital and analog Peak to peak transmit voltage at the MDI port 7/13/2004 2

3 Transmit waveform and peak voltage test Modeled after Test Mode 1 and 4 in 1000BASE-T PHY transmits the specified sequence of PAM data symbols, similar as in Gigabit Ethernet standard, Example outputs in test modes 1 and 4 in 1000BASE-T : 7/13/2004 3

4 Key specs in the waveform test mode Voltage level at the MDI for a totally random pattern through the transmitter, modeled after the test mode 4 of 1000BASE- T (primarily for excluding Baseline Wander) = 2V peak to peak differential. Should take into account the transformer insertion loss, as transformer is a part of the transmitter. Level accuracy proposal, in test mode 1 = +/- 0.44dB (~ +/- 5% ) instead of +/- 0.83dB (~ +/- 10%) in 1000BASE-T» 10GBASE-T requires more predictable voltage levels, so as to not reduce already low system margin. Level matching accuracy of the PAM symbols in test mode 1 = +/- 1% as in 1000BASE-T.» Same as in 1000BASE-T, clause /13/2004 4

5 Transmitter droop test Depends primarily on the transformer high pass pole. Backward compatibility to Gigabit requires lower than 100kHz high pass pole (informative rather than normative) 128 identical symbols sent for the droop test (<160ns for Fs>800MHz) Maximum droop defined over 64 symbol periods (<80ns for Fs > 800MHz)= 5.5% 1000BASE-T defined maximum droop over 64 symbols as ~27% for Fs=125MHz Reduces the overhead of design on transmitter and receiver due to baseline wander to much lower than 1000BASE-T. 7/13/2004 5

6 For defining waveform template Filter assumptions specified as information to create the templates, only templates being normative. Analog filter assumptions proposal: A two (or more) pole continuous time filter will exist One pole comes from the transformer One from the output capacitance of the transmitter (bond-wire inductance and output impedance of the transmitter) Proposal: -3dB frequency varying from fs/2 to ¾ fs, fs being the sampling rate. Single pole continuous time high pass filter with pole 100kHz to model the lower -3dB frequency of the transformer. 7/13/2004 6

7 Jitter test modes and spec. Test setup similar to 1000BASE-T test mode 2 and 3 where the transmitter sends alternating +M,-M symbols timed from a clock with frequency Fs +/-0.01% in the MASTER and SLAVE timing mode respectively. In 1000BASE-T: Peak to peak jitter of the differential signal at the MDI w.r.t. TX_CLK called J txout, wherein TX_CLK is the transmitter output clock. Jitter on TX_CLK filtered w.r.t. a high pass filter with pole of ~5kHz in Master device and 32kHz in the slave device in 1000BASE-T. 1000BASE-T specification for filtered jitter plus J txout was 300ps for Master mode (test mode 2) and 400ps more than the Master in Slave mode (test mode 3) In 10GBASE-T, peak to peak value of the jitter on the differential signal at the MDI and on the TX_CLK, to be less than 30ps peak to peak. The peak-to-peak value of jitter measured over 10 6 clock edges (~ +/- 5σ) Measurement was done over 10^5 clock edges in 1000BASE-T 7/13/2004 7

8 Test fixture for waveform/jitter tests Transmitter under test 100 ohms High impedance diff probe Digital Oscilloscope or data acquisition module Test Fixture 1:Transmitter test fixture for template and waveform measurement Transmitter under test 100 ohms High impedance diff probe TX_CLK Jitter Analyzer Test Fixture 2: Transmitter test fixture for jitter measurement 7/13/2004 8

9 TX nonlinearity modeling and measurement Transmitter sending real TX data X + α1*x 2 +α2*x 3 Channel X + α1*x 2 +α2*x 3 Transmitter (far end) Receiver (near end) with Linear Echo canceller Receiver (far end) with Linear Echo canceller Nonlinearity of the far end and local transmitter modeled the same. Only second and third order nonlinearity modeled. Second order nonlinearity due to differential nature of circuits will be much lower than third order in actual practice. Determine the coefficients of transmit second and third order nonlinearity in a real system to meet 10G net rate. Nonlinearity/distortion spec tight due to the limited echo cancellation in the local receiver for a linear echo canceller. 7/13/2004 9

10 TX nonlinearity time-domain modeling Transmitter sending real TX data Receiver (near end) with Linear Echo Canceller X + α1*x 2 +α2*x ohm V d = 0.8V at fs/8 Far end transmitter and cable replaced with a sinusoidal disturber with series 100ohm impedance to produce similar rate of 10Gbps for same α1 and α2 as determined in previous slide. Determine amplitude V d and frequency to have similar effect on capacity and net rate as the real far end signal. V d determined to be 0.8V peak to peak differential so that at MDI it corresponds to 0.4V peak to peak differential, and average frequency of ~fs/8. 7/13/

11 TX distortion time-domain measurement test setup 50 ohm Transmitter under test V d 50 ohm High impedance diff probe Digital Oscilloscope or data acquisition module Post-Processing Purity of the signal source V d such that all harmonics at least 75dB below fundamental. Obtained by band pass/low pass filtering the output of a signal source. Post processing block removes the disturbing signal, by removing the best fit sine wave at the fundamental frequency of the disturber. Distortion spec captures also the device noise of the transmitter, which is equally important for own local receiver performance. 7/13/

12 Time-domain measurement challenges Availability of Oscilloscope and high impedance probe with low noise and low jitter. Example scope and probe: Agilent series and Infinimax probes with 25kohm impedance With averaging the noise might be acceptable, and by the available datasheets, the result almost bit accurate. Low jitter: Scope jitter needs to be less than ~1ps rms. Limited accuracy in time-domain measurement because of the resolution of the ADC generally in the measurement equipment. Suitable equipment to be investigated. 7/13/

13 Transmit time-domain distortion test Specified in presence of a disturber with 0.4V peak to peak differential amplitude, at a frequency of fs/8, with transmitter transmitting a sequence of 2047 symbols Number of symbols same as in 1000BASE-T Disturber voltage reduced as justified in prior slide. Peak distortion measured by sampling the differential output w.r.t. any arbitrary phase and processing a block of 2047 consecutive samples with the MATLAB code from Gigabit Ethernet modified to get to the required accuracy (modified from clause , IEEE std , section three) Removes sinusoidal disturber & does a minimum mean square error fit Measured with test fixture 3, and processed through the MATLAB code, the peak distortion output of the code proposed to be less than TX dist mv. 7/13/

14 TX nonlinearity in frequency domain Transmitter sending CW with freq. fin, same peak amplitude X + α1*x 2 +α2*x 3 Spectrum analyzer For same α1 and α2 as in Step 1, measure ratio of Signal power to net harmonic power. This was reported in slide 5 of presentation Easier to measure than time-domain approach. 7/13/

15 Frequency domain measurement of TX nonlinearity More availability of spectrum analyzers to capture the DAC output. If the pre-coder approach is taken, the transmitter can output a sine wave, and all the harmonics measured on the spectrum analyzer. Scott Powell, Kevin Chan, Broadcom, Pvt. conversation Approximate Specs in this mode discussed in May meeting of 802.3an task force. The transmit nonlinearity to be measured under the condition of a single tone sine wave of peak-peak amplitude of 2V, of frequencies ~ fs/16, fs/8, & fs/4 and measuring all the harmonics to be below the fundamental by ( TX nonlin ) db. AND A two tone output with a peak to peak amplitude of 2V, the two tones being close to fs/2, with spacing less than 1MHz, and all the inter-modulation components being measured below the fundamental tones by at least ( TX nonlin + 2.5) db. Limitation: Does not accurately capture the device noise contribution of transmitter to uncancelable echo. 7/13/

16 Distortion test for 10GBASE-T The measurement method of distortion to be either one of the previous tests, i.e. either in the time domain through the test setup described in slide 11, and detailed in slide 12-13, OR in the frequency domain specified and detailed in slide Observation: Nonlinearity/distortion spec is tight due to the own local receiver performance. Note: Nonlinearity/distortion spec could be specified as loose to meet the 10G performance of the far end device only. In that case, risk of under-specifying and misleading a local transmitter design which may not meet 10G performance on local receiver, and break the link, even being spec compliant. 7/13/

17 Distortion specification: Far end/near end Because of reasoning in the last bullet point of last slide: Concept of normative spec/recommended spec to be mentioned in the standard (Scott Powell, Broadcom, Pvt. Conversation): Specify a normative spec for interoperability of the far end receiver. a recommended spec. for the near end receiver. The numbers discussed in the initial proposals in the previous slides can be considered mainly for recommended specifications for near end receiver. Recommended specification can be TX dist 1.5mV, TX nonlin ~ 68dB. Normative spec. to be met with respect to a frequency mask. At low frequency ( 25MHz), the linearity spec similar to the one proposed for recommended spec, as the TX nonlinearity appears similarly at both the far end and near end receiver. For margin over different link segment lengths, initial proposal for parameter TX nonlin in the range of (65,70) db for 0-25MHz, and roll of in the range of [(65,70) 20*log10(f/25)] db, f 25, f in MHz. Exact values in the above equation TBD. (Theoretically linearity mask can even follow something close to IL curve) Linearity measured w.r.t. the spectrum analyzer approach as detailed in slide 15. Sine wave outputs for frequency < fs/4, Two tone outputs for IM tests for high frequencies. time domain measurement approach does not address frequency dependent nonlinearity spec. 7/13/

18 MDI impedance balance Specified in 1000BASE-T with the following test setup: In 1000BASE-T, over a frequency range of 1-100MHz, the common mode-to-differential impedance balance of each channel of MDI, specified to exceed *log10(f/50) db, f in MHz. The balance being defined as 20*log10(E cm /E diff ) In 10GBASE-T, the frequency range needed is 1-625MHz. The specification for common mode-to-differential balance for 10GBASE-T based on measurements as suggested in Baseline Text for Clause 55.X MDI specifications, Terry Cobbs, Systimax, or TBD. 7/13/

19 MDI common mode output voltage 1000BASE-T spec: Magnitude of the common mode output voltage V cmout, on any transmit circuit, when measured as shown in the test setup shown below (reproduced from IEEE Std , Clause , section three), shall be below 50mV peak to peak, when transmitting data. Specification for 10GBASE-T to be kept TBD 7/13/

20 Common mode noise rejection 1000BASE-T provided a common mode rejection test where you inject a common mode signal at any frequency from 1-250MHz through a cable clamp, with the output of the signal generator adjusted for a voltage of 1.414V peak on the oscilloscope. Cable clamp test specified in clause and in Annex 40B of IEEE std /13/

21 10GBASE-T common mode rejection test Proposed changes to above for 10GBASE-T Increase frequency range: 1-625MHz. Common mode signal imposed on the 10GBASE-T device, should take into account the limited CMR of the transformer magnetic at high frequencies Pulse transformer CMR plots show 6dB/octave roll-off for frequencies greater than 100MHz. Recommendation: Cable clamp specification to be widened up-to 625MHz, and should be such that the common mode output signal at the receiver input should be 2.8V for frequencies from 1-80MHz. 2.8V*80/f for frequencies > 80MHz, f in MHz. Such applied voltage should be representative of real environment specified. To be determined by the following tests: For Europe for ITE including LAN products: EN55024 For central office equipment : GR1089 in the USA and EN in EU 7/13/

22 Alien Crosstalk Noise rejection Alien Crosstalk noise rejection test very important for qualification of the 10GBASE-T device. 1000BASE-T assumed ANEXT coupling transfer function was spectrally flat. 25mV peak to peak noise source imposed on the MDI using an external 100BASE-T transmitter coupled with a resistive divider. Clause , figure /13/

23 Alien Crosstalk Noise rejection proposal In 10GBASE-T, ANEXT test should be performed with some model of real ANEXT coupling built in the test. A real 10GBASE-T transmitter driving 100 ohm produces the Power Spectral Density of the transmitter. Alien crosstalk for 10GBASE-T, CAT6, channel model #3 is specified as: *log10(f/100), f in MHz for f > 100MHz *log10(f/100), f in MHz for f < 100MHz. Alien crosstalk for 10GBASE-T, CAT6, channel model #2 (55m CAT6) is specified as: *log10(f/100), f in MHz for f > 100MHz *log10(f/100), f in MHz for f < 100MHz. Need two different networks for reproducing test setup modeling the two different ANEXT channel models. 7/13/

24 ANEXT noise rejection proposal The slope of 10*log10(f) or 15*log10(f) difficult to reproduce in a test setup (different than a single pole high pass filter) Channel model #3: An approximation to both the ANEXT voltage(~ 5mV peak to peak differential) and its spectral shape is obtained by passing the real PSD of a asynchronous 10GBASE-T transmitter through a 10:1 attenuator and a simple high pass filter. Attenuator of 10:1 needed for reasonably large value of coupling capacitors as shown in next figure coupled from the network to the victim receiver (with differential impedance of 50ohm) Channel model #2: 2:1 attenuator used for similar order coupling capacitors to the victim receiver. Capacity and net rate achieved for the real ANEXT models in both Channel model #2 and #3 and the test network very close 7/13/

25 Proposed Alien Crosstalk noise rejection test setup (for Channel model #3, ANEXT= *log10(f/100) ) T1 MDI Link Segment MDI R1 Transmitter (4 Twisted pairs) T2 Receiver (4 twisted pairs) R2 T3 R3 T4 R4 200 ohms 90 ohms 20 ohms 90 ohms 300fF 300fF Noise source: Independent 10GBASE-T transmitter sending PAM data asynchronous to the 7/13/ GBASE-T transmitter under test applied to each pair simultaneously. 25

26 ANEXT and proposed network s ANEXT (for Channel model #3, ANEXT= *log10(f/100) ) 40 Comparison of PSD of ANEXT and the proposed test filter Proposed filter PSD ANEXT PSD( log(f/100) 140 Comparison of net Alien NEXT and proposed test Alien NEXT PSD of ANEXT, proposed filter (in db) Net Alien NEXT, and proposed test ANEXT Proposed net ANEXT Net ANEXT (Tx PSD + ANEXT PSD) Frequency (MHz) Frequency (MHz) 7/13/

27 Proposed Alien Crosstalk noise rejection test setup (for Channel model #2, 55m Class E ANEXT= *log10(f/100)) T1 MDI Link Segment MDI R1 Transmitter (4 Twisted pairs) T2 Receiver (4 twisted pairs) R2 T3 R3 T4 R4 200 ohms 50 ohms 100 ohms 50 ohms 350fF 350fF Noise source: Independent 10GBASE-T transmitter sending PAM data asynchronous to the 7/13/ GBASE-T transmitter under test applied to each pair simultaneously. 27

28 ANEXT and proposed network s ANEXT (for Channel model #2, 55m Class E, ANEXT= *log10(f/100) ) 30 Comparison of PSD of ANEXT and the proposed test filter 120 Comparison of net Alien NEXT and proposed test Alien NEXT Proposed net ANEXT Net ANEXT (Tx PSD + ANEXT PSD) PSD of ANEXT, proposed filter (in db) Proposed filter PSD ANEXT PSD( log(f/100) Net Alien NEXT, and proposed test ANEXT Frequency (MHz) Frequency (MHz) 7/13/